When defendants speak: Quantifying the predictive value of defence arguments in construction litigation

OpenAlex · English

The global construction industry faces significant risks due to disputes.This study aims to predict outcomes in construction dispute judicial decisions by analyzing the linguistic interaction between plaintiff claims and defendant defenses in Turkish, addressing a methodological gap in the literature.The research examines 2,563 Court of Cassation decisions in Türkiye from 2011-2021 (from 15,667 cases), organized into three datasets: containing both plaintiff claims and defendant defenses (Dataset I), only plaintiff claims (Dataset II), and all decisions (Dataset III).Dataset I uniquely captures the impact of defendant voice, demonstrating how including counterarguments significantly enhances model performance.Standard preprocessing techniques were applied to address Turkish morphological challenges.Among various feature extraction methods, TF-IDF demonstrated superior performance.The HistGradientBoosting achieved optimal performance, with Dataset I reaching 87.38% accuracy compared to 84.53% for Dataset II, proving that modeling mutual arguments enhances prediction beyond using plaintiff claims alone, exceeding success rates in comparable literature.This study pioneers a framework for analyzing the dialectics of legal texts in construction disputes, with applications across different legal systems.